Journal of Educational Psychology

Practicing Versus Inventing With Contrasting Cases: The

Effects of Telling First on Learning and Transfer
Daniel L. Schwartz, Catherine C. Chase, Marily A. Oppezzo, and Doris B. Chin
Online First Publication, August 22, 2011. doi: 10.1037/a0025140

CITATION
Schwartz, D. L., Chase, C. C., Oppezzo, M. A., & Chin, D. B. (2011, August 22). Practicing
Versus Inventing With Contrasting Cases: The Effects of Telling First on Learning and
Transfer. Journal of Educational Psychology. Advance online publication. doi:
10.1037/a0025140
Journal of Educational Psychology © 2011 American Psychological Association
2011, Vol. ●●, No. ●, 000 – 000 0022-0663/11/$12.00 DOI: 10.1037/a0025140

Practicing Versus Inventing With Contrasting Cases:

The Effects of Telling First on Learning and Transfer

Daniel L. Schwartz, Catherine C. Chase, Marily A. Oppezzo, and Doris B. Chin

Stanford University

Being told procedures and concepts before problem solving can inadvertently undermine the learning of
deep structures in physics. If students do not learn the underlying structure of physical phenomena, they
will exhibit poor transfer. Two studies on teaching physics to adolescents compared the effects of
“telling” students before and after problem solving. In Experiment 1 (N ⫽ 128), students in a tell-and-
practice condition were told the relevant concepts and formulas (e.g., density) before practicing on a set
of contrasting cases for each lesson. Students in an invent-with-contrasting-cases (ICC) condition had to
invent formulas using the same cases and were told only afterward. Both groups exhibited equal
proficiency at using the formulas on word problems. However, ICC students better learned the ratio
structure of the physical phenomena and transferred more frequently to semantically unrelated topics that
also had a ratio structure (e.g., spring constant). Experiment 2 (N ⫽ 120) clarified the sources of the
effects while showing that ICC benefited both low- and high-achieving students.

Keywords: transfer, science education, contrasting cases, inventing, proportional reasoning

In a review of the Third International Mathematics and Science
Study, Hiebert and Stigler (2004) noted that instruction in the
United States largely takes a form that we will label tell and
practice (T&P). Teachers or texts first explain concepts and their
formulaic expression, and then students practice on a set of well-
designed problems. It is a convenient and efficient way to deliver
accumulated knowledge. Nevertheless, many scholars are working
on instructional alternatives. Catrambone (1998) summarized a
prevailing concern with T&P: “Students tend to memorize the
details of how the equations are filled out rather than learning the
deeper, conceptual knowledge” (p. 356).
Many T&P alternatives use some form of guided discovery,
including problem-based learning (Barrows & Tamblyn, 1976),
project- and design-based activities (Barron et al., 1998), inquiry
(Edelson, Gordin, & Pea, 1999), and modeling (Lesh & Doeer,
2003). The mechanics of these alternatives withhold didactic
teaching at first, lest it undermine the processes of discovery.
Daniel L. Schwartz, Catherine C. Chase, and Marily A. Oppezzo,
School of Education, Stanford University; Doris B. Chin, Human-Sciences
and Technologies Advanced Research Institute, Stanford University.
This material is based upon work supported by the National Science
Foundation under grants SMA-0835854 & DRL-1020362. Any opinions,
findings, and conclusions or recommendations expressed in this material
are those of the authors and do not necessarily reflect the views of the
National Science Foundation. The authors thank Kimi Schmidt for help in
executing the empirical study; Dylan Arena for his help in the survey of the
transfer literature; Melissa Gresalfi for her help in designing and imple-
menting the video protocol; and, Carl Wieman for encouraging us to
demonstrate the point.
Correspondence concerning this article should be addressed to Doris B.
Chin, Human-Sciences and Technologies Advanced Research Institute,
Stanford University, 450 Serra Mall, Building 160, Stanford, CA 94305-
2055. E-mail: dbchin@stanford.edu
However, there are unresolved debates on the timing of explicit
instruction (see Kirschner, Sweller, & Clark, 2006; Tobias &
Duffy, 2009). Sweller (1988), for instance, proposed that with-
holding explicit instruction incurs a needless cognitive load that
detracts from learning. In contrast, Bransford, Franks, Sherwood,
and Vye (1989) proposed that students first need to experience the
problems that render told knowledge useful. The persistence of
these debates has been attributed to the lack of experimental
evidence (Mayer, 2009). One reason for the lack of evidence is that
T&P and guided discovery can differ on so many dimensions that
it is difficult to maintain fidelity to their respective instructional
models while isolating the causal differences between them.
For example, Schwartz and Bransford (1998) asked students to
invent graphs to characterize simplified data sets from classic
psychology experiments. The students learned and transferred
more from a subsequent lecture than other students who first
summarized an explicit chapter on the experiments and then heard
the same lecture. Schwartz and Bransford argued that there is a
“time for telling.” For novices, explicit instruction is effective if
they have been prepared with appropriate experiences. But if they
have not been prepared, then explicit instruction provides much
less value. Supporting this claim, Schwartz and Martin (2004)
found that students who invented a variance formula for a set of
specially designed cases exhibited superior transfer from a subse-
quent worked example, as compared with students who completed
T&P instruction before the worked example. In response to these
findings, Sweller (as cited in Schwartz, Lindgren, & Lewis, 2009,
p. 53) correctly questioned
the appropriateness of the control groups used in these studies. If
multiple factors are varied simultaneously, as they were in these
experiments, does this procedure not break the “vary one thing at a
time” rule essential to all randomized, controlled experiments, result-
ing in it being impossible to determine exactly what caused the
effects?

1
2 SCHWARTZ, CHASE, OPPEZZO, AND CHIN
In other words, it is impossible to know whether the results were
due to the specially designed cases, the timing of explicit instruc-
tion, or both.
The following studies highlight a guided discovery pedagogy
called inventing with contrasting cases (ICC). The contrasting
cases are designed to help students find “deep structure” (we say
more below). The contrasting cases can also serve as practice
problems for T&P. Therefore, they provide an opportunity to
maintain both instructional fidelity and experimental control. In
the following two studies, all conditions used the same sets of
contrasting cases. The principal difference between conditions was
that T&P students were told the concepts and solution methods
beforehand and then practiced with the materials. Students in the
ICC condition invented their own solutions for the same materials
and were not told about the conventional solutions and concepts
until the end. Given the close comparison, the current experiments
begin to identify the specific effects of telling first on subsequent
learning and transfer compared with one form of guided discovery.
Inventing With Contrasting Cases
ICC is similar to other instructional approaches that ask students
to generate symbolic representations or models of data (e.g., Bow-
ers, Cobb, & McClain, 1999; Lehrer & Schauble, 2004; Lesh &
Doerr, 2003). The unique aspects of ICC are that the inventing
tasks take a few minutes rather than days or weeks, and they use
predefined contrasting cases. We start by explaining the contrast-
ing cases. Afterward, we describe the inventing activity.
Contrasting Cases
Contrasting cases, which originally came from work in percep-
tual learning (J. J. Gibson & Gibson, 1955), are instructional
materials designed to help students notice information they might
otherwise overlook. As with tasting wines side by side, contrasts
can improve discernment (e.g., Biederman & Shiffrar, 1987;
Bransford et al., 1989; Eisner, 1972). Rittle-Johnson and Star
(2007, 2009), for example, found that students developed more
procedural flexibility in algebraic manipulations if they compared
alternative solutions side by side than if they saw them sequen-
tially. By contrasting different solutions, students noticed what
made each procedure unique and useful.
In addition to supporting procedural flexibility with symbolic
representations, contrasting cases can be configured to highlight
empirical regularities. Figure 1 shows a set of contrasting cases
designed to help students learn about density. The narrative of the
worksheet is that each row represents a company that ships clowns
to events. A given company always packs its clowns into buses by
the same amount, but each company packs its clowns to different
degrees.
The systematic variability across the cases is intended to help
students notice the ratio structure of density. Gick and Paterson
(1992), for example, demonstrated that the inclusion of near
misses— contrasts that differ on a single dimension—improves
schema induction and transfer. In Figure 1, the buses in the upper-
and lower-right corners have the same number of clowns but vary
in size (number of bus compartments). This should help students
notice that bus size is important and not just the total number of
clowns.
The cases in Figure 1 contain at least three levels of structure.
The first level is the surface features, or irrelevant structure. Two
examples of surface features are the type of clown and the lines
delineating the bus exterior. These incidental details are irrelevant
to the concept of density. If this were the only information students
gained from the cases, they would not have learned much about
density.
A second level of structure is the density used within a given
company. For example, the company in the first row uses two
buses with different numbers of clowns and bus sizes, but both
buses use a density of one clown per compartment. This level of
information better instantiates the idea of deep structure, because
the density is a fixed relation between mass and volume, even
though the amounts of mass and volume vary.
The third level is the structure of ratio as it occurs across the
companies. Although the specific densities differ for each com-
pany, all three use proportionate ratios. This last level of structure
is termed the invariant under transformation (J. J. Gibson, 1979).
The invariant of a ratio structure persists in all the cases, despite
the varying ratios for each company.
An invariant under transformation is a specific class of deep
structure that differs from the qualitative relations (Forbus, 1984)
often investigated in research on analogical transfer (e.g., an atom
is like the solar system; Gentner & Markman, 1997). For this class
of deep structure, there is a lawful parametric relation (ratio) that
must be preserved across changes to the quantities, and a given
ratio is defined by its position along a continuum of possible ratios.
The contrasting cases in Figure 1 are designed on the assumption
that picking up the invariant of a ratio structure across situations is
the key to effective transfer and that without this recognition, it
seems unlikely students would transfer to other ratio phenomena
such as speed or springiness.
Inventing
Students need a productive task orientation to benefit from the
contrasting cases. The literature provides several good examples.
Williams and Lombrozo (2010) proposed orienting students with
an “explanation” directive. In their research, participants received
artificial stimuli that comprised contrasting cases of category in-
stances. Participants who were asked to explain category member-
ship exhibited better learning and transfer compared with thinking
aloud or simply describing the instances. Similarly, Schmidt, De
Volder, De Grave, Moust, and Patel (1989) asked students to work
in small groups to explain the contrasting case of a blood cell that
expands in water but shrinks in saltwater. These explanation ac-
tivities prepared students to learn from a relevant text compared
with students who had not done the group explanation activities.
Taking a different approach, Gentner, Loewenstein, and Thomp-
son (2003) found that students who received directives to draw
analogies across negotiation strategies transferred more than stu-
dents who did not.
In ICC, students are asked to invent a quantitative index. For the
cases in Figure 1, students invent an index for density, disguised as
a “crowdedness index” for each company. They have to come up
with a single, common way to compute a consumer index that
works for each company, so that consumers can compare the
crowdedness used by each company (because crowded clowns are
grumpy clowns). The directive to make an indexing scheme differs
 COMPARING USES OF CONTRASTING CASES 3

Figure 1. Crowded clowns worksheet. Students in the tell-and-practice condition were told the formula for
density and then practiced applying it to determine the crowdedness of the clowns for each company. Students
in the invent-with-contrasting-cases condition were not told about density but instead had to invent their own
crowdedness index for each company.

from asking students to explain, draw an analogy, or simply
compare and contrast, because it specifically pushes students to-
ward a single, quantitative explanation of the deep structure.
Ideally, the inventing task also recruits a confluence of productive
psychological processes including explanation, analogy, and com-
pare and contrast. One question addressed by the current research
is whether inventing is a useful task orientation or whether it
would be simpler and more direct just to tell students about density
and have them practice with the cases.
Learning Predictions
In debates over the timing of explicit instruction, studies that
have found advantages for “telling first” have focused on the
learning of multistep procedures. For example, Klahr and Nigam
(2004) found an advantage for explicit instruction over unsup-
ported discovery for the control-of-variable procedure. Tuovinen
and Sweller (1999) found that worked examples improved data-
base programming compared with free exploration, especially for
novices. Although unsupported exploration has low fidelity to
most models of guided discovery, these results indicate that ex-
plicit instruction can improve procedural knowledge. However,
success at procedural learning does not address a major challenge
for many topics of instruction. In science, students also need to
learn the structure of scientific phenomena.
Our hypothesis is that explicit instruction beforehand may not
lead students to learn the deep structure very effectively. When
learning science, students often focus on the surface features of
single instances rather than the deep structures that define a phe-
nomenon across instances (e.g., Goldstone & Son, 2005). For
example, students may hear about density and learn its formula,
and then facilely map the variables of the formula to discrete
features of the subsequent problems. In short, they may simply
4 SCHWARTZ, CHASE, OPPEZZO, AND CHIN
divide the relevant values to compute density. Though successful,
they may not understand the significance of ratio in the structure of
density. Thus, we are making the possibly unintuitive prediction
that telling students the structure of a phenomenon, in this case
with a verbal explanation and an encapsulating mathematical for-
mula, may limit their learning of the structure from subsequent
practice problems.
Initial support for this hypothesis comes from the previously
mentioned time-for-telling studies (Schwartz & Bransford, 1998;
Schwartz & Martin, 2004). Kapur (2008, 2010) also compared
“delay of structure” treatments with T&P instruction in a series of
studies on Newtonian physics and rate–time– distance problems. In
the delay of structure treatments, students did not initially receive
explicit instruction that ensured they could solve the problems. The
students often failed, but it was a “productive failure,” because the
students exhibited better overall learning. One interpretation is that
the T&P students, who were told how to solve the problems
beforehand, had no need to search for the deep structure. In
contrast, the delay-of-structure condition may have led students to
search for the deep structure so they could solve the problems.
To gather more direct evidence on the effects of T&P for the
learning of structure, the following studies took a simple measure-
ment approach. Students practiced or invented with a set of con-
trasting cases, and a day later they redrew the cases. Our first
prediction was that T&P students would not reproduce the ratio
structure of the problems as well as ICC students who received the
same problems without being told first.
A second prediction was that T&P students would transfer
poorly. Gick and Holyoak (1983) demonstrated that if people do
not learn the deep structure, they rarely exhibit spontaneous trans-
fer to problem isomorphs with different surface features. There-
fore, we predicted that at posttest, T&P students would transfer
relatively poorly to scientific phenomena that involve ratio but not
density (e.g., spring constant), despite being able to solve proce-
dural problems on density. People can have knowledge of a math-
ematical formula, or any formal theory, without recognizing the
structure it describes. For example, Michael, Klee, Bransford, and
Warren (1993) showed that clinical psychology students could
recite the relevant theories but could not apply those theories to
patients. They had not learned to recognize the structure of symp-
toms in patients.
Experiment 1
In Experiment 1, we examined eighth-grade students’ learning
of structure and their spontaneous transfer from lessons on density
and speed to problems on surface pressure and the spring constant.
In terms of physical phenomena, these four are quite different.
Nevertheless, they share the deep structure of a ratio between
distinct physical properties. Density is described as mass over
volume, speed is distance over time, surface pressure is force over
area, and the spring constant is force over displacement.
Ratio is a critical deep structure for many physical phenomena.
By adolescence most children have the cognitive wherewithal to
reason about ratio (Siegler, 1981), but this does not mean they
spontaneously recognize or learn the ratio structure in novel situ-
ations. If students do not pick up the ratio structure during instruc-
tion, they should do poorly on transfer problems that have a ratio
structure. In contrast, helping students understand the prevalence
of the ratio structure should open up a deeper understanding of
many physics concepts.
Method
Participants. Participants (N ⫽ 128) were eighth graders
from four classes at a diverse middle school (35% Asian, 25%
Latino, 22% Filipino, 11% White, 4% African American, and 3%
other; 37% qualified for free-lunch programs). Within each class,
students were assigned to the T&P or ICC condition through
stratified random assignment based on their cumulative class
scores in science. Treatments occurred in separate rooms with
rotating instructors to ensure there were no teacher effects (each
instructor taught both T&P and ICC for each lesson). In both
conditions, students worked in self-elected and sometimes chang-
ing pairs, consistent with regular class practice. All tests were
taken individually.
Design and instruction. This section describes the complete
treatments for both conditions. The experiment had two phases that
served to test the two predictions. The first phase tested the
prediction that T&P students would not learn the structure of
density as well as ICC students. The second phase of the study
tested the prediction that T&P does not support transfer as well as
ICC instruction. Figure 2 shows the full timeline for instruction
throughout the study. Measures of student learning were inter-
leaved at four points (shown centered and bolded).
Phase 1. On Day 1, T&P students received an instructional
page titled “Finding Density” (see Figure 3A). It contained some
everyday examples, the formula, and a worked example that was
different from the contrasting cases they would receive. Students
took turns reading sections of the page aloud, while the rest of the
class followed along. Student questions were answered by refer-
ring them to the instruction page. Afterward, students were told,
“On the next page, compute the density for each company that
buses clowns.” Students then practiced on the contrasting cases
worksheet in Figure 1. The instructional page was kept by students
for reference.
In the ICC condition, students did not receive any density
instruction beforehand, but instead they received an instruction
page titled “Inventing an Index” (see Figure 3B). It described an
index in everyday language and gave several examples, such as
class grade and batting average. It developed the narrative of compa-
nies shipping clowns and the need for a crowdedness index. The
instructional page did not provide any examples for how to compute
an index, but it provided a set of constraints at the end of the page (i.e.,
a company only gets one index value, the same procedure should be
used to find the index for each company, and a larger index means
more crowded). The instructional page was read aloud, and questions
were answered in the same fashion as in the T&P condition. Then,
students were told to “invent a procedure for computing a crowded
clown index for each company.” They used the same contrasting
cases worksheet that the T&P group received. The instructional page
was kept by students for reference.
The critical measurement for Phase 1 came on Day 2. Twenty-
four hours after completing the clowns worksheet, and prior to any
further potentially contaminating instruction or experiences, stu-
dents were asked to reconstruct the worksheet. This memory test
provided a measure of the structure they had learned from the
previous day’s lesson (Chase & Simon, 1973; Jee, Gentner, For-
 COMPARING USES OF CONTRASTING CASES
Figure 2. Design of Experiment 1. The tell-and-practice (T&P) students
received identical practice cases and tests as the invent-with-contrasting-
cases (ICC) students. T&P students were told key formulas and concepts,
and then practiced on the cases. ICC students had to invent an index for
each set of cases without being told the appropriate formulas. The images
provide snippets of the materials students received. Picture recall, word
problem, transfer, and delayed transfer tests served as measures of learning
and transfer.
bus, Sageman, & Uttal, 2009). Students received a sheet that
stated: “Yesterday, you received a sheet that had an activity with
clowns and buses. Use the space to redraw how the sheet looked
the best you can.” Students were given 10 min to complete their
drawings. We coded two aspects of their recreations. The first was
whether students reconstructed the ratio structures. We call this
deep structure recall. Students received 1 deep structure point for
each pair of buses that shared one ratio between them, unique from
the other pairs of buses (3 points ⫽ 100%). The ratios did not have
to be the same as in the original worksheet, but they had to be
proportional across pairs of buses. The second coding captured
students’ memory for surface features. Students received 1 point
for each of the following surface features that we had a priori
seeded into the cases (6 points ⫽ 100%): elaborated clown fea-
tures, clowns positioned on lines between bus compartments,
different line styles for the buses, wheels that did not correspond
to the compartments, a company name, and incidental text features
such as “Name ___.” Our prediction was that surface feature recall
would be the same in both conditions but that structural recall
would be worse in the T&P condition. This would indicate that the
5
effect of T&P is specific to deep structure. A primary coder
evaluated this and all other measures, and a secondary coder
evaluated 25% of the data for each measure. Measures were
blind-coded, and agreement was greater than 98% for each mea-
sure for both studies.
Phase 2. Students completed three more lessons using T&P
or ICC on Days 2 and 5. The lessons were designed to cover
discrete and continuous versions of both density and speed. (The
full set of materials is available at http://aaalab.stanford.edu/
transfer.html.) As in Day 1, the T&P condition always received
and read aloud an instructional page that included everyday ex-
amples of the target concept, an introduction to the relevant for-
mula, and a worked example that used a different cover story from
the contrasting cases. The ICC instructions also kept to their
format from Day 1. Students were asked to invent an index for the
contrasting cases. However, it was no longer necessary to explain
the concept of inventing an index or introduce the constraints on
the index (e.g., each company only gets a single index score). After
their respective instructional pages, students received the worksheets,
which had the same data and contrasting cases regardless of condition.
The insets of Figure 2 show one case from each of these worksheets.
The popcorn task had students find the popping rate of different
machines (discrete speed). The gold cube task had students find the
purity of gold used by different companies (continuous density). The
race car task had students find the speeds of different cars and, further,
determine which cars have the same speed (continuous speed) and
therefore come from the same company. (Note that this additional
instruction for the cars task requests the T&P and ICC students to
explicitly compare across cases for the first time.)
In addition to the instructional pages, a second treatment differ-
ence involved the timing of an explicit lecture on ratio (shown in
bold italics in Figure 2). T&P students received a lecture on the
importance of ratio in physics at the beginning of Phase 2, on Day
2. This lecture was separate from the instructional pages the T&P
students received before each of the remaining three lessons. The
lecture directly explained the prevalence of ratio in physics includ-
ing density, speed, and force, and it explicitly indicated the anal-
ogous ratio structures in the clowns worksheet, an approach dem-
onstrated to support transfer (Brown & Kane, 1988; Gentner et al.,
2003; Gick & Holyoak, 1983). The lecture helped to evaluate the
common intuition that if one just told the students that ratio was
important and explicitly pointed it out, they would begin to learn
the structural information from the subsequent contrasting cases
and use it later at transfer.1 ICC students received this same
1
A reasonable question is why the overarching lecture on ratio was
given after the T&P students had worked on the clowns worksheet and not
before. We did not want to use the lecture before students completed Phase
1 because the lecture would cloud the strict comparison of T&P instruction
versus ICC instruction on the redrawing measure. If the T&P students had
received a lecture and then an introductory lesson specific to density before
the memory measure, one might argue that a broad orienting lecture is not
typical of T&P instruction. Alternatively, one might argue that the T&P
students were overloaded by the extra instruction, which might hinder their
learning from the subsequent density lesson with the clowns. Therefore, we
avoided using the lecture in Phase 1. To determine whether the broader
ratio lecture overcame problems with T&P, we can look at the transfer
measures that also test whether students had learned the deep structure of
ratio, though less directly than the redrawing measure.
6 SCHWARTZ, CHASE, OPPEZZO, AND CHIN

Figure 3. Instructional pages for crowded clowns worksheet. Instructional information was read aloud by
students. (A) Tell-and-practice instructions contained an introduction to the concept with everyday examples, the
relevant formula, and a worked example. Students were then told, “On the next page, compute the density for
each company that buses clowns.” (B) Invent-with-contrasting-cases instructions explained the concept of an
index with everyday examples, provided a narrative for finding a crowdedness index, and introduced rules for
the index. Students were then told, “Invent a procedure for computing a crowded clown index for each
company.”

lecture, but on Day 8, after they had completed all the inventing
tasks. The lecture also served as the ICC students’ explicit instruc-
tion on the concepts of density and speed.
The last piece of instruction for both groups was a worksheet of
word problems, given on Day 8. This gave students practice
applying the density and speed formulas. Across the 4 days of
instruction, the ICC instruction required approximately 10 more
minutes of instructional time. The T&P students used this time
completing extra word problems on this last day.
To test the effects of T&P and ICC on transfer, three learning
measures were taken during Phase 2. The first, an immediate
transfer task, was given at the beginning of Day 8, after ICC
students had completed the four ICC cycles but had not yet
received the ratio lecture. This permitted us to evaluate transfer for
the ICC students before they had received any direct instruction,
whereas the T&P students had completed all their instruction
(except practice on word problems). Students received a sheet that
showed diagrams of four aerosol cans with internal springs press-
ing on plates. Students were asked to describe the plate pressure
used in the aerosol cans. The word describe had not been used
previously in either treatment. Students received 1 point for each
instance that was described with a ratio (e.g., springs over plate
area; 4 points ⫽ 100%).
The second measure was a word problem test given on Day 11.
The test comprised three problems each for density and speed. An
example problem is “Brenda packs 120 marshmallows into 4 soda
cans. Sandra packs 300 marshmallows into 11 soda cans. Whose
soda cans are more densely packed?” This test served two pur-
poses. First, as is characteristic in transfer research, it is important
to ensure that a lack of transfer is not due to a simple failure to
learn the concept or procedure. Therefore, we wanted to be sure
the T&P students had learned how to compute ratio quantities and
could do so in the context of solving standard word problems. The
second purpose was to find out whether the ICC condition put
these students at risk for not learning how to apply the formulas,
given that they did not practice on the contrasting cases and
received much less overall direct instruction on the density and
speed concepts and their formulas.
The third Phase 2 measure was a delayed transfer task that
occurred 21 days after all students had completed their instruc-
tional treatments. Ideally, this delay would determine whether the
effects of the treatments were somewhat lasting, and it would
determine whether the final lecture helped or hurt the ICC stu-
dents. All students attempted the transfer problem embedded in a
biology test. Figure 4 shows that the transfer problem was on the
spring constant. Students were asked to “determine the stiffness of
 COMPARING USES OF CONTRASTING CASES
Figure 4. Transfer task on spring constant. Students were asked to
determine the stiffness of the trampoline fabrics, which is the equivalent of
finding their spring constants. Though a far semantic transfer from speed
and density, it shares the deep structure of ratio.
the mat fabric for each trampoline.” Students received 1 point for
each trampoline described by ratio (e.g., number of people over
displacement; 4 points ⫽ 100%).
Results
Redrawing test: Encoding of surface features and deep
structure. Figure 5 shows excerpts of student recreations of the
clowns worksheet. The examples show what surface and deep
feature recall looks like for this task. Table 1 shows the average
percentages of recall for deep structures and surface features in
Experiment 1. The T&P students did not recreate the deep struc-
ture as well as the ICC students. There was a strong interaction
between treatment and the type of feature recalled, F(1, 120) ⫽
11.9, p ⬍ .001. There was a treatment effect for deep structure,
F(1, 120) ⫽ 11.0, p ⫽ .001, d ⫽ 0.60, but not for surface features,
F(1, 120) ⫽ 1.5, p ⫽ .223, d ⫽ 0.22. There were no significant
correlations between surface and deep recall (rICC ⫽ ⫺.21, p ⫽
.10; rT&P ⫽ .09, p ⫽ .52). Interestingly, 19% of the T&P students
wrote down the formula, which was not present on the original
worksheet, compared with only 3% of the ICC students. Recall of
the ratio structure did not differ appreciably for those T&P stu-
dents who wrote down the formula (M ⫽ 57%, SD ⫽ 33.6%)
versus those who did not (M ⫽ 51%, SD ⫽ 43.9%), F(1, 56) ⫽
0.2, p ⫽ .65. This result highlights that remembering the structure
of the formula does not entail knowing the structure of the phe-
nomena to which the formula refers.
Transfer effects. By the first transfer task on surface pres-
sure, T&P students had received lessons on discrete and continu-
ous versions of density and speed, and they had received a separate
lecture on the importance of ratio in physics that explicitly mapped
7
the analogy across density, speed, and other physics concepts. For
each lesson, they had received a unique worked example and then
practiced on a new set of cases. The means in Table 1 indicate that
the T&P students used ratio to describe surface pressure signifi-
cantly less often than the ICC students, who had not yet been told
about the formulas or the importance of ratio, F(1, 94) ⫽ 6.8, p ⫽
.011, d ⫽ 0.53. (Due to an implementation error, data from both of
the conditions in one class could not be included in the preceding
analysis of immediate transfer. The delayed transfer analysis,
which is next, included all the students.)
Figure 6 provides prototypical examples of how students an-
swered the delayed transfer problem on the spring constant (tram-
poline). (The immediate transfer problem on surface pressure had
the same classes of response.) To receive credit for transfer,
students had to produce answers similar to the one in Figure 6D,
which indicates a ratio between people and rungs. The delayed
transfer problem occurred 3 weeks after both groups had com-
pleted the final word problems and the ICC students had received
their lecture on ratio. Table 1 indicates that the ICC students
performed significantly better than the T&P students on the de-
layed transfer test, F(1, 124) ⫽ 13.6, p ⬍ .001, d ⫽ 0.66.
Further analysis indicated that learning the ratio structure during
the clown task was important for transfer, regardless of condition
(see Klahr & Nigam, 2004, on common causes of transfer despite
instructional differences). A three-step regression analysis used
performance on the delayed transfer problem (trampoline) as the
dependent measure. In the first step, structural recall of the clown
task was regressed against transfer performance, F(1, 119) ⫽ 5.79,
p ⫽ .018, r ⫽ .22. Students who recalled the deep structure of the
clowns were more likely to transfer. In the second step of the
analysis, experimental condition was added to the regression equa-
tion. It improved the prediction of the model, Fchange(1, 118) ⫽
8.84, p ⫽ .004, rmodel ⫽ .34. Thus, students in the ICC condition
were more likely to transfer, even when taking into account their
structural recall of the clowns worksheet. One possible explanation
is that ICC students who did not induce the ratio structure during
the clown activity learned it over the three subsequent inventing
activities. Finally, in the third step, the interaction term of condi-
tion by clown recall was added to the equation. It did not improve
the fit, Fchange(1, 117) ⫽ 0.91, p ⫽ .342, rmodel ⫽ .35. The lack of
an interaction indicates that finding the clown structure improved
transfer for both conditions. The primary difference was that
students in the ICC condition found the structure more often.
Figure 5. Samples of clown recall. Samples of students’ worksheet
recreations, partitioned by the inclusion of surface and deep features.
8 SCHWARTZ, CHASE, OPPEZZO, AND CHIN

Table 1
Means (in Percent) of Outcome Measures by Experiment and Treatment, Including a Further Breakout of Experiment 2 Based on a
Median Split of Achievement Using Students’ Prior Class Performance

Experiment 1 Experiment 2

Broken out by prior achievement

ICC T&P ICC T&P ICC T&Pa

Measure M SD M SD M SD M SD High M Low M High M Low M

Deep structures 75.5ⴱⴱ 35.3 52.3 41.9 58.6ⴱⴱ 40.6 38.4 36.0 73.1 46.9 48.1 23.2
Surface features 27.1 18.9 31.3 19.0 33.1 14.8 28.5 15.2 35.3 31.2 29.2 27.5
Ratios at transfer surface pressure 35.2ⴱ 46.2 13.8 32.9
Ratios at transfer spring constant 53.5ⴱⴱ 50.0 23.0 42.4 50.9ⴱⴱ 49.9 23.3 41.8 64.0 40.9 32.9 8.7
Word problems 70.1 23.1 75.2 20.3 64.8 19.6 66.4 19.5 75.0 57.0 72.9 56.5

Note. ICC ⫽ invent with contrasting cases; T&P ⫽ tell and practice.
a
Two T&P students did not have achievement scores.
ⴱ
Comparison of treatment means, p ⬍ .05. ⴱⴱ Comparison of treatment means, p ⬍ .005.

Overall, these results should not be interpreted as demonstrating
that finding the structure of the clowns worksheet was sufficient
for transfer. A more cautious interpretation is that those who found
the structure on the first lesson were more likely to find structure
on the second, third, and fourth lessons, which would have a
cumulative effect for improving transfer.
One potential concern with the focus on ratio is that students
may have been learning math, but they may not have been learning
physics. A post hoc coding of the transfer problems sheds light on
this concern. We coded whether students ranked the trampoline
cases correctly in terms of their physical property of stiffness. The
instructions did not have an explicit directive to rank the cases, but
the students implicitly ranked them, for example, by writing “stiff-
est,” or a specific ratio, next to a trampoline. Students who used
ratios for over half the cases correctly ranked the relative stiffness
of the fabrics 96% of the time. Students who did not use ratios
succeeded 1% of the time, ␹2(1, N ⫽ 126) ⫽ 113.7, p ⬍ .001. (The
results are similar for the surface pressure problem, ␹2(1, N ⫽
96) ⫽ 66.1, p ⬍ .001.) Students who did not use ratios tended to
rank the stiffness of the fabric based on the value of a single
feature such as the number of people on the trampoline (see Figure
6A) or the number of rungs the fabric stretched (see Figure 6B).
Thus, without ratios, students had very little chance of forming a
qualitative understanding of the physics, because they focused on
only one feature in the physical situation. As an analogy, students
will often conflate density and mass, because they do not appre-
ciate that density is a structural relation between two physical
properties.
Word problem test: Computation. The transfer differences
were not due to differences in students’ knowledge of the formu-

Figure 6. Sample solutions for the trampoline problem. Students’ solutions for delayed transfer captured the
total number of people (A), the total number of rungs (B), a global impression of stretchiness (C), or the ratio
structure of rungs by people (D). Only the latter was considered an instance of transfer.
 COMPARING USES OF CONTRASTING CASES
las. Table 1 indicates that the T&P and ICC conditions exhibited
similar percentages correct on the word problem test, F(1, 124) ⫽
1.7, p ⫽ .20.
Discussion
The T&P and ICC students had learned to apply formulas to
word problems equally well. Nevertheless, T&P instruction was
suboptimal for helping students learn the structures that are im-
portant for transfer. T&P students exhibited a relatively low rate of
recreating the deep structure of the contrasting cases at recall, and
they also transferred less often than the ICC students, both before
and after the ICC students had received a lecture on ratio in
physics. Our favored explanation is that many T&P students fo-
cused on applying what they had been told, and missed the deep
structure in the problem situations.
The T&P students’ poor recall of the ratio structure of the
crowded clowns cannot easily be attributed to a lack of interest or
general memory effects. If the T&P results were due to general
effects, then the T&P students’ memory for both surface features
and deep structures should have been lower than that of the ICC
students, but the T&P and ICC students demonstrated equal sur-
face feature recall.
Regardless of condition, students who found and recalled the
ratio structure of the crowded clowns worksheet transferred better.
However, more of the ICC students found the ratio structure, and
therefore more ICC students transferred. Under this explanation,
the ICC students’ deeper understanding of ratio led to the transfer.
An alternative explanation is that the ICC students had “learned to
learn” (e.g., Brown & Kane, 1988). They developed the strategy of
looking across multiple problems to find a deep structure. By this
alternative, the ICC students did not transfer the ability to recog-
nize the ratio structure, but instead they transferred the idea of
looking for patterns given multiple cases. The transfer problems
included multiple cases that may have cued the same strategy they
had used during instruction. If true, it means that ICC students
should not transfer the use of ratios to situations that only involve
a single instance. Experiment 2 tests this prediction.
Experiment 2
What led to superior levels of transfer for the ICC students?
Experiment 2 was designed to help answer this question by exam-
ining (a) which behaviors helped students learn the structural
information and (b) what learning led to the transfer. We assume
that the contrasting cases themselves were an important element
that aided students in learning the deep structure. But they could
not have been sufficient because the T&P students also used them.
Our hypothesis is that the inventing directive orients students to
search the cases for a common deep structure on which to base
their index. In contrast, T&P instruction focuses students on ap-
plying solutions, one problem at a time. This reduces their chances
of finding the deep structure across the cases. To investigate
further, we videotaped a subset of students. We coded the video-
tapes specifically for how many times the students transitioned
among the cases. The prediction was that students in the ICC
condition would shuttle among the cases as they searched for the
structure on which to base the index. In contrast, the T&P students
would be more inclined to complete each case separately and
9
sequentially, making fewer transitions between cases. As Gentner
et al. (2003) concluded from their studies on analogical encoding,
“Learners cannot be counted on to spontaneously draw appropriate
comparisons, even when the two cases are presented in close
juxtaposition” (p. 403).
The second issue is the question of what transferred. One
hypothesis is that students transferred the ability to recognize
ratios. An alternative hypothesis is that students transferred the
strategy of looking across multiple cases. To test these alternatives
at transfer, we gave half the students from each condition the
original four-case trampoline problem, and the other half received
the same scenario but with only one trampoline. If the ICC
students transfer on the basis of having a problem with multiple
cases, like the problems at instruction, then the ICC advantage
over T&P should disappear for the one-case transfer problem. If
the ICC students transfer on the basis of ratio, then they should
exhibit greater transfer than T&P students for both the one- and
four-case transfer problems.
A final practical question addressed by Experiment 2 is whether
ICC and T&P instruction interacts with students’ prior achieve-
ment. There are concerns that lower achieving students may be
better served by the greater guidance of T&P instruction. For
example, Hiebert and Wearne (1996) found that early elementary
students with low initial understanding did not fare very well with
invention tasks when learning place value. Similarly, Tuovinen
and Sweller (1999) compared adult learning through free explora-
tion with learning from worked examples. Free exploration was
effective only for adults with high prior knowledge. Therefore, we
collected permission to analyze students’ cumulative class grades
in science to see whether it would indicate that the T&P instruction
was relatively more effective for the low-achieving students.
Method
Experiment 2 followed a nearly identical protocol with 120
eighth-grade students at the same school. There were four primary
changes: (a) The experiment used instructors blind to the hypoth-
eses; (b) 24 pairs of students were selected at random to be
videotaped in class while they worked on the clowns worksheet (3
pairs ⫻ 4 classes ⫻ 2 treatments); (c) the timeline was condensed
by removing the first transfer task on surface pressure and com-
bining the word problem and delayed transfer test into a single
posttest, given 1 week after instruction; and (d) students in each
treatment received either the one- or four-case trampoline transfer
problem (spring constant) at random.
Results
Table 1 shows the results for Experiment 2. The data replicated
Experiment 1 closely. There were no differences between treat-
ments for surface feature recall and word problem performance,
but the ICC condition showed a substantial advantage for deep
structure recall and transfer performance. Although science
achievement had an overall effect on performance, it did not
interact with the instructional conditions. The right-most columns
show the Experiment 2 data partitioned by a median split on the
students’ cumulative grades in their science class. The treatment
effect was sufficiently strong that lower achieving ICC students
10 SCHWARTZ, CHASE, OPPEZZO, AND CHIN

descriptively outperformed higher achieving T&P students at
transfer, 40.9% versus 32.9%.
To statistically dissect the possible interactions of treatment by
achievement, we performed separate analyses of covariance on the
four measures in Table 1. The analyses used the achievement
covariate, the treatment factor, and the Treatment ⫻ Achievement
interaction. Two students in the T&P condition did not have
cumulative grades and were omitted from these analyses.
For deep structure recall, ICC significantly outperformed T&P,
F(1, 111) ⫽ 8.7, MSE ⫽ 0.12, p ⫽ .004, d ⫽ 0.53; achievement
was a significant predictor of deep structure recall, F(1, 111) ⫽
23.7, p ⬍ .001, but there was no Treatment ⫻ Achievement
interaction, F(1, 111) ⫽ 0.15, p ⫽ .70. For surface feature mem-
ory, there was no difference by treatment, F(1, 111) ⫽ 2.8, MSE ⫽
0.02, p ⫽ .10, d ⫽ 0.31; no effect of achievement, F(1, 111) ⫽ 1.6,
p ⫽ .22; and no interaction, F(1, 111) ⫽ 0.09, p ⫽ .77. As in
Experiment 1, there were no significant correlations between sur-
face feature and deep structure recall (rICC ⫽ .09, p ⫽ .49; rT&P ⫽
⫺.18, p ⫽ .17).
With respect to posttest performance, the ICC students exhibited
greater transfer, F(1, 111) ⫽ 10.6, MSE ⫽ 0.20, p ⫽ .002, d ⫽
0.60; there was an effect of achievement on transfer, F(1, 111) ⫽
9.2, p ⫽ .003, but there was no Treatment ⫻ Achievement inter-
action, F(1, 111) ⫽ 0.18, p ⫽ .67. For the word problems, there
were no treatment differences, F(1, 111) ⫽ 0.21, MSE ⫽ 0.03, p ⫽
.65, d ⫽ 0.08; there was a very strong effect of achievement, F(1,
111) ⫽ 34.0, p ⬍ .001, but again there was no interaction of
treatment by achievement, F(1, 111) ⫽ 0.09, p ⫽ .76.
In this study, we also coded the percentage of correct answers
students produced for each of the four contrasting cases work-
sheets they received during instruction (these data were not avail-
able for Experiment 1). The T&P condition did modestly better.
An analysis of variance compared the average percentage correct
per worksheet and demonstrated a significant effect of condition
(MICC ⫽ 84.7%, SD ⫽ 22.7%; MT&P ⫽ 91.8%, SD ⫽ 12.2%), F(1,
113) ⫽ 4.5, p ⫽ .037. The following disaggregates worksheet
performance across the four lessons: clowns (discrete density;
MICC ⫽ 85.1%, SD ⫽ 31.7%; MT&P ⫽ 91.5%, SD ⫽ 23.6%), F(1,
113) ⫽ 1.52, p ⫽ .22; popcorn (discrete speed; MICC ⫽ 82.7%,
SD ⫽ 31.8%; MT&P ⫽ 90.4%, SD ⫽ 23.2%), F(1, 113) ⫽ 2.19,
p ⫽ .14; gold (continuous density; MICC ⫽ 85.7%, SD ⫽ 33.5%;
MT&P ⫽ 96.6%, SD ⫽ 13.4%), F(1, 113) ⫽ 5.33, p ⫽ .023; race
car (continuous speed; MICC ⫽ 85.1%, SD ⫽ 29.1%; MT&P ⫽
88.7%, SD ⫽ 28.8%), F(1, 113) ⫽ 0.44, p ⫽ .51. In isolation, only
the gold worksheet exhibited a significant treatment difference.
This is driven by an uptick in T&P performance, which one may
speculate is the result of the worksheet’s overt numerical presen-
tation of weight and volume, which in turn made it easier for the
T&P students to map in the density formula. In sum, despite doing
somewhat worse on the basic classroom assignments, the ICC
students did better at transfer and equally well on the word prob-
lem test. In absolute terms, the ICC students exhibited a high
degree of success at the inventing task.
What learning transferred? Figure 7 indicates a consistent ad-
vantage for the ICC condition for both one- and four-case prob-
lems, F(1, 112) ⫽ 10.8, p ⫽ .001. Although the four-case problem
was solved more frequently than the one-case problem, the differ-
ence was not significant, F(1, 112) ⫽ 2.7, p ⫽ .105. Importantly,
there was no Treatment ⫻ Number of Cases interaction, F(1,
112) ⫽ 0.5, p ⫽ .481, which indicates that the ICC students

Figure 7. Transfer performance by treatment and number of cases included in the problem. The advantage for
invent with contrasting cases (ICC) over tell and practice (T&P) on both one- and four-case problem types
indicates that students were transferring on the basis of picking up the ratio structure and not on a strategy of
searching for structure across multiple cases.
 COMPARING USES OF CONTRASTING CASES
transferred their understanding of ratio. If ICC students had trans-
ferred only a strategy of inventing across multiple cases, they
would not have shown a similar advantage over the T&P condition
for the one- and four-case versions. Notably, the T&P students also
benefited from the four-case version, which makes sense, given
that the T&P students were confronting a set of contrasting cases
without being told a formula beforehand, similar to the ICC
students’ original instruction.
As in Experiment 1, regardless of treatment, students who
recalled the deep structure of the clowns worksheet were more
likely to transfer than those who did not. A three-step regression
analysis used the performance on the transfer problem (trampo-
line) as the dependent measure. In the first step, structural recall of
the clown task was regressed against transfer performance, F(1,
111) ⫽ 14.37, p ⬍ .001, r ⫽ .34. Students who recalled the deep
structure of the clowns were more likely to transfer. In the second
step of the regression analysis, experimental condition was added
to the regression equation. It improved the fit of the model,
Fchange(1, 110) ⫽ 5.13, p ⫽ .025, rmodel ⫽ .39. Overall, students
in the ICC condition were more likely to transfer, presumably
because they noticed the ratio structure on the subsequent invent-
ing tasks. Finally, in the third step, the interaction term of condi-
tion by structural recall was added to the equation. It did not
improve the model’s fit, Fchange(1, 109) ⫽ 0.76, p ⫽ .385, rmodel ⫽
.40. The benefit of finding the structure of the clowns for transfer
did not differ by condition. It was just that the ICC students learned
that structure more often.
One possible source of the ICC advantage is that students
searched the contrasting cases to find a deep structure on which to
base their index. To find out whether the ICC students searched
across the cases more, we coded the frequency with which each
videotaped pair transitioned their attention from one clown com-
pany (row) to another. Pointing to, writing on, or discussing a
particular company was coded as a “reference” to that case. A
transition occurred when at least one student of the pair shifted
from referencing one company to another. The minimum number
of transitions is two, one between each of the three companies. The
ICC pairs transitioned an average of 20.1 times (SD ⫽ 10.3),
compared with 6.0 (SD ⫽ 4.7) for the T&P pairs, F(1, 22) ⫽ 18.7,
p ⬍ .001, d ⫽ 1.76. As with the full sample, the videotaped ICC
pairs recalled the ratio structure of the clowns more than the
videotaped T&P pairs (average of pair recall: MICC ⫽ 81.3%,
SD ⫽ 33.0%; MT&P ⫽ 40.3%, SD ⫽ 29.7%), F(1, 22) ⫽ 4.71, p ⫽
.041.
T&P students already had the formula, and they tended to apply
it to each case separately, which reduced their chances of noticing
the invariant of ratio across the cases. Within the T&P condition,
nine of 10 pairs exhibited 10 or fewer transitions, and there was no
correlation between number of transitions and a pair’s average
structural recall, r(12) ⫽ .05, p ⫽ .87. For the ICC condition, eight
of the 10 pairs showed more than 10 transitions. This suggests that
these eight ICC pairs were searching the cases to find the invariant
property on which to base their index. For ICC, there was a
nonsignificant negative correlation between number of transitions
and structural recall, r(12) ⫽ ⫺.25, p ⫽ .38. Past a certain
threshold, more transitions did not yield better learning. For ex-
ample, the ICC pair with the most transitions (42) kept searching
the cases because they never discovered the invariant structure. At
recall, they only recreated one ratio.
11
The videotapes were collected to test a specific behavioral
prediction (number of transitions) rather than generate a corpus for
protocol analysis. However, to enrich the picture of the process in
action, we provide protocol excerpts from a pair of students from
each treatment as they found their first answer for one of the clown
companies. It should be noted that these excerpts were specifically
chosen to highlight treatment differences. The preceding statistical
analysis shows that they reflect general differences, although there
was natural variation across the pairs. With each transcript, we
provide a brief commentary pointing out the key features from our
perspective. For clarity, companies are referred to as Company A,
B, or C, from top to bottom on the Figure 1 worksheet.
T&P pair. The following transcript reflects the T&P stu-
dents’ tendency to focus on the formula such that the task becomes
a problem of mapping the variables of the formula to the discrete
features of a specific case. Notice that these students begin by
explicitly stating the formula “D equals M over V” and then work
from there. These students, like others, spend a good deal of time
figuring out that each compartment (rather than a complete bus)
represents one unit of volume. Moreover, these students make only
one transition—from Company C to Company B—throughout the
entire segment.
Student 1: Okay, this one kinda makes more sense [points to Company
C]. Okay, this one, I think, is 2 objects. There’s 6 things in here, so . . .
Student 2: Or are there 3?
Student 1: Oh, okay, I see, I see . . . I’m gonna start with this one first
[Company C].
Student 2: Let me write it. D equals M over V. [writes on sheet]
Student 1: Okay. Well . . . I’m just going off how . . . what she
[the instructor] did up there. So there’s 2 objects there [points to
Company C] . . . and add.
Student 2: Do we put this in there? [points to Company C answer line]
Student 1: No, I’m just doing it [the compartments] together, rather
than separate. So, hold on here.
Student 2: The mass is how many things are inside of it.
Student 1: So together there’s 6, individually there’s 4. So I think . . .
Okay, it’s for each bus [compartment].
Student 2: So this [compartment] is separate?
Student 1: Think so? So each one . . . are separate ones, yeah.
Student 2: But why is it dashed line?
Student 1: I have no clue. The same thing right here. Hmm, well. So
I don’t think . . .
Student 2: [points to Company B] This one 1, 2, 3, 4, 5, 6, 7, 8, 9, and
3 buses [compartments] plus 1 bus [compartment] is 4, so 12, so 12
divided by . . . 4 is 3.
Student 1: There you go! So 3 here.
Student 2: Three objects per cube.
ICC pair. The following transcript, of about the same length
from an ICC pair, illustrates how they made far more transitions as
they looked across cases to notice critical features and tried to
develop an indexing scheme.
Student 3: So, we’re writing down a number?
Student 4: Yeah, we’re writing down a number, a value.
12 SCHWARTZ, CHASE, OPPEZZO, AND CHIN
Student 3: Do we count the clowns, or do we count the buses? ’Cause
this one’s separate from the other ones, I think . . .
Student 4: I think . . . Lemme read the . . . [flips to instructions then
back to worksheet]. Okay, so it’s not about the number of clowns, it’s
about how crowded it is. Which one do you think is the least crowded?
Student 3: [points to Company A] It’s a bigger bus.
Student 4: The most crowded?
Student 3: [points to Company C, then Company B] One of these two?
Student 4: I think it’s this one. [points to Company C]
Student 3: Because, if these two get added in, they’re really crowded
[points to Company C], and this one too. [points to Company B]
Student 4: Let’s see. 1, 2, 3, 4 — 4 in 2 sections [points to Company
C]. 1, 2, 3, 4, 5, 6, 7, 8, 9. [points to Company B]
Student 3: There’s 3. [points to Company B]
Student 4: There’s 3 in each section [points to Company B], this is 2
in each section [points to Company C], so this seems more crowded
[points to Company B]. So what do you think we should do?
Student 3: Should we add these guys in here? [points to Company C]
Student 4: Um, no, I think that’s good. I think this is less [points to
Company A], this is most [points to Company B], this is middle.
[points to Company C]
The group has a total of 12 transitions within this segment of
conversation, reflecting how they moved back and forth from one
case to another. It is useful to note that the transitions were in the
service of finding a single account. They were not simply com-
paring and contrasting the cases, but rather they were driving
toward a single mathematical explanation of crowdedness. In
comparing companies to figure out which was the most crowded,
the students realized that they could not discriminate the crowd-
edness level between Companies B and C without counting. It was
only after counting the clowns and bus compartments that they
decided Company B was the most crowded. Finally, they checked
their indexing procedure by their intuitive sense of crowdedness.
Their quantitative activity and their intuitive sense of crowdedness
worked together to sharpen what they learned from the cases.
Discussion
As in Experiment 1, the T&P and ICC groups performed sim-
ilarly on the word problem test and remembered similar numbers
of surface features on the clown recall task. The ICC group
recalled the density structure better and exhibited higher rates of
transfer. Performance on the clown recall task was a good predic-
tor of success on the delayed transfer task, regardless of condition.
The ICC group was just more likely to understand the structure by
the first day of instruction. Interestingly, the ICC group was
slightly less successful than the T&P group in solving the con-
trasting cases, indicating that initial rates of success at the learning
activity do not necessarily predict later transfer rates (cf. Kapur,
2008).
What transferred? The ICC advantage over T&P did not in-
crease for the four-case problem. Instead, the ICC students main-
tained the same relative advantage for the single-case problem.
This makes it likely that they transferred an understanding of the
ratio structure, rather than being dependent on a problem with
multiple cases.
How did students come to understand the underlying structure in
the first place? The video analysis demonstrated that students in
the ICC condition made far more transitions between the cases as
they searched for common structure. The T&P students, on the
other hand, went through the cases in a linear fashion, mapping the
formula to each company in turn.
The relative effects of the ICC instruction compared with T&P
did not differ systematically for individuals of different levels of
achievement. Low-achieving students benefited from ICC com-
pared with T&P as much as high-achieving students. It is useful to
note that students were working in pairs, so it is still possible that
working individually would lead to a prior Treatment ⫻ Achieve-
ment interaction.
General Discussion
Standard T&P instruction is important because it delivers the
explanations and solutions invented by experts, and students need
opportunities to hear and practice these ideas. For students to gain
this benefit without undermining transfer, the current studies sug-
gest that expositions should happen after students have explored
novel deep structures. The inventing activity can serve as “prep-
aration for future learning” by readying students to appreciate
more fully the expert solutions and deep structures when they
are explained (Bransford & Schwartz, 1999). Giving students the
end-product of expertise too soon short-cuts the need to find the
deep structure that the expertise describes. Students in the T&P
condition focused on what they had been told, and they applied the
formulas sequentially to the problems, which reduced their
chances of finding the deep structure. Without an appreciation of
deep structure, students are less likely to see the structure in new
situations that differ on the surface, and they will fail to transfer.
The current studies tried to address the challenge of maintaining
instructional fidelity to different pedagogies while isolating a
causal variable. The difficulty of doing both simultaneously may
be one reason why much of the transfer research is predicated on
comparing T&P instruction to itself. This makes it possible to vary
only one thing, while maintaining fidelity to the dominant model
of instruction. We sampled transfer articles of math and/or science
learning published between 2003 and 2008.2 Among the 70 articles
(136 unique authors) that resulted from our search, 75% of the
studies used T&P instruction for both treatment and control con-
ditions while investigators manipulated other variables. T&P in-
struction was not a variable in these studies and therefore not
suspected as a potential contributor to transfer success or failure.
As the current studies show, the model of instruction is a large
contributor to transfer, and therefore many of the psychological
claims based on transfer studies may not generalize beyond T&P
instruction. For example, consider the relation between abstract
and concrete elements of instruction. Some scholars favor abstract
presentations to avoid obscuring problem structure (e.g., Bassok &
2
An ISI Web of Knowledge search used transfer conjoined with think-
ing, procedure, learning, or psychology. The resulting 350 articles were
winnowed to those in English that reported original data on the transfer of
math and/or science learning. For each study, two coders separately eval-
uated whether a solution procedure or concept was provided prior to
students’ independent problem solving. The 3% disagreements were re-
solved by discussion.
 COMPARING USES OF CONTRASTING CASES
Holyoak, 1989; Harp & Mayer, 1998; Kaminski, Sloutsky, &
Heckler, 2008), whereas others favor concrete instances to connect
to prior knowledge (Goldstone & Sakamoto, 2003; McNeil,
Uttal, Jarvin, & Sternberg, 2009). Many of the relevant studies,
however, have used T&P instruction for both conditions. By
hypothesis, T&P instruction does not help students pick up the
deep structure, so it makes sense that surface features would
appear to be an issue in many studies. In the current studies, if
surface features interfered with picking up deep structure, then
there should have been a negative correlation between surface
feature and deep structure recall (e.g., Rothkopf & Billington,
1979), but there was not. Moreover, Schwartz, Chase, and Brans-
ford (in press) had students work with either the crowded clowns
worksheet or an analogous abstract worksheet that used dots in
cubes. For T&P instruction, the abstract worksheet led to better
structural recall, consistent with other studies that have used T&P
instruction (e.g., Bassok & Holyoak, 1989). However, for the ICC
instruction, the students performed the same for either version of
the worksheet, and the ICC students doubled the rate of structural
recall found for the T&P students who received the abstract
worksheet. These examples demonstrate the risk of generalizing
psychological claims (e.g., about the value of abstract examples)
without taking into consideration the broader instructional context.
Alternative Hypotheses
On the relatively low performance of T&P. Our execution
of the telling portion of T&P was consistent with prevalent class-
room practice (Hiebert & Stigler, 2004), although we fortified the
approach with multiple worked examples, practice across far anal-
ogies, and a lecture that explicitly described the importance of ratio
and further pointed out the ratio analogy across separate physical
domains. Our goal was to maximize the reach of the findings
across current educational practices by demonstrating a natural
psychological response to being told the form of an answer before
working on problems. Nevertheless, it is possible that our results
will not generalize across all variants of direct instruction (see
Atkinson, Derry, Renkl, & Wortham, 2000, for a review). For
instance, direct instruction may be relatively effective for teaching
discrete facts or procedural steps, whereas density involves a
principled relation between dimensions (volume and mass). More
pertinent to the current studies, it seems reasonable that if the
teachers had shown the general ratio structure in the clowns
worksheet before students had worked on it, the students would
have exhibited better recall of the deep structure.
Showing students the ratio structure may improve structural
encoding for the specific problems, but it may not transfer as well
as having students induce the structure on their own. In the current
studies, T&P students were explicitly told and shown the ratio
structures in the lecture after the clowns worksheet. However, they
did not transfer very well, despite having three more lessons that
depended on ratios. Perhaps T&P students were less able to rec-
ognize the ratio structure on the transfer problems because they
never had the experience of actively recognizing the ratio structure
on their own (Roll, 2009). Another possibility is that students in
typical classrooms, compared with laboratory settings, may not
allocate critical lessons on structure the special attention they
deserve. They may be more interested in just learning how to solve
the problem.
13
With regard to the practice portion of T&P, the task of comput-
ing answers to problems is fairly typical, and students had further
practice on subsequent word problems. The formulas for density
and speed are relatively straightforward equations. The T&P stu-
dents were able to apply them effectively to the T&P cases and
posttest. This leads to the possibility that the practice was so easy
for the T&P students that they did not have to do very deep
processing, and therefore they learned less. Perhaps T&P becomes
more effective when tasks become more difficult. Before embrac-
ing this hypothesis fully, it is important to note that the T&P
students engaged in a good deal of germane cognitive effort as
they tried to map the quantities to the variables of the equation (see
transcript in Experiment 2).
A second issue is that telling first may be more effective for
learning complex procedures, though less effective for complex
structures. The current studies do not bear on this hypothesis
except in demonstrating that early explicit instruction in science,
which is designed to help students handle procedural complexities,
runs the risk of focusing student attention on the procedures
themselves at the expense of the structures that the procedures
were designed to encapsulate. There are alternatives to T&P for
teaching complex procedures. Schwartz and Martin (2004), for
example, demonstrated that an ICC sequence better prepared stu-
dents to understand the complex formulas for statistical variance
compared with procedural instruction at the outset.
On the relatively high performance of ICC. These studies
were not meant to isolate the causal effect of each ingredient in
ICC instruction. Here we offer some speculation about these
ingredients, plus predictions about possible studies to test those
ingredients. We assume that the contrasting cases are one crucial
component because they can help students find structure in phe-
nomena. Without these specially designed cases, which highlight
critical similarities and differences, it would be difficult to find the
underlying structure. A simple study might compare the outcomes
of inventing for the current crowded clowns worksheet (see Figure
1) versus a worksheet where all the companies use the same ratio
(with no built-in contrasts). Our prediction is that the latter con-
dition would not fare as well.
Another key ingredient is the behavior triggered by the directive
to come up with a single account, or index, for all the cases. The
protocols of student problem solving showed that ICC students
revisited each company about seven times on average. In the
process, the students were actively comparing and contrasting the
cases as they transitioned back and forth between them.
The benefits of comparison for transfer have been well docu-
mented (e.g., Gentner et al., 2003), and it is presumably necessary
for the effect of ICC. But it may not be sufficient. A useful study
would compare how students perform in a compare-and-contrast
condition versus an ICC condition. Our prediction is that if stu-
dents were simply instructed to compare and contrast the cases
without driving toward a single parametric explanation, the stu-
dents would compare and contrast familiar features rather than find
the underlying structure that makes apparent differences the same.
A final ingredient for the current instance of ICC is the quan-
titative information that helped students to learn the ratio structure
with greater precision. Without countable quantities, the students
would probably make comparisons with terms like more and less,
which lack the precision required to find the ratio structure.
Schwartz, Martin, and Pfaffman (2005), for example, found that
14 SCHWARTZ, CHASE, OPPEZZO, AND CHIN
children who were asked to give verbal explanations to balance
scale problems did not learn to relate the dimensions of weight and
distance. In contrast, children who were asked to invent mathe-
matics to explain their answers were more likely to discover that
proportionate ratios determine whether the scale balances. Without
the precision and demand of quantitative reasoning, students may
gloss over important specifics and possible relations. For instance,
we would predict that students would not learn as well if they were
asked to explain the companies instead of inventing a quantitative
index.
The brief inductive activities of the ICC physics lessons did not
cover many important aspects of density and speed. One concern
is that the mathematical focus of inventing may have been at the
expense of qualitative understanding. Our assumption, however, is
that helping students learn the deep structure prepares them to
transfer to learn more about these concepts later. For instance,
understanding density as a ratio of mass to volume should help
students understand buoyancy, where density, mass, and volume
are frequently conflated. Thus, a useful study would determine
whether this is true.
The ICC instruction did not improve strategic knowledge for
finding patterns across multiple instances, as demonstrated by the
lack of a special benefit for the four-case transfer problem over the
one-case problem. However, the current implementation did not
have the time to emphasize the development of strategic knowl-
edge and scientific dispositions (Gresalfi & Cobb, 2006). Taylor,
Smith, van Stolk, and Spiegelman (2010) found that students who
received a full course of inventing activities in college-level cell
biology were more able to generate explanations for novel cell
phenomena compared with T&P students.
Conclusions
Practical applications of ICC. There are different types of
learning that range from skill acquisition to identity formation, and
it seems unlikely that a single pedagogy or psychological mecha-
nism will prove optimal for all types of learning. ICC is one among
many possible ways to support students in learning deep structure.
Other alternatives to T&P include problem-, project-, and inquiry-
based instruction, as well as approaches that attempt to “problema-
tize” tasks so that students will not take the eventual solutions for
granted (Bransford et al., 1989; Fensham & Kass, 1988; Hiebert et
al., 1996; Limón, 2001; Needham & Begg, 1991). All these ap-
proaches will support productive transfer to the extent they help
students find the deep structure that generalizes across situations.
The ICC activities differ from many of these pedagogies, however,
because ICC strategically precedes standard T&P pedagogy rather
than replaces it. Moreover, the ICC activities do not sacrifice
overall classroom efficiency for the sake of transfer. The ICC
treatment required about 10 extra minutes of instruction across 4
days, which was mostly spent on explaining the novel task of
inventing an index.
The inventing task differs from discovery tasks because students
do not have to rediscover the answer discovered by experts.
According to Experiment 2, students actually did reinvent versions
of the true formulas over 80% of the time. But other work has
demonstrated that if students notice a subset of the relevant struc-
ture, this is sufficient to help them make sense of the rest, given a
formal exposition (Kapur, 2008; Schwartz & Bransford, 1998).
The contrasting cases are designed to support incremental, even
partial, induction rather than one-shot insight. Thus, when used in
classrooms, it is important to help students tolerate the short-term
ambiguity of not being told the right answer. The effort to find and
characterize the structure can improve learning and test perfor-
mance in the long run.
Toward a theoretical account. Cognitive psychology theo-
ries provide one way to describe the mechanisms behind the
effects of ICC. For example, one might propose that the inventing
activities helped students use analogical processes to abstract a
ratio schema and this abstract schema enabled the transfer (e.g.,
Gick & Holyoak, 1983). An alternative class of theories starts with
perception (e.g., Greeno, Smith, & Moore, 1993; Hofstadter,
1995). In particular, J. J. Gibson’s (1979) ecological theories of
perception and perceptual learning provided useful guidance for
developing the predictions, measures, and contrasting cases. Al-
though the ICC tasks engage cognitive mechanisms, and therefore
should not be reduced to purely perceptual mechanisms, many of
the insights from perceptual learning apply.
With respect to predictions and measures, a perceptual learning
account proposes that students need to learn to pick up or notice
information in the environment (E. J. Gibson, 1969). For example,
expert radiologists can see diagnostic details in X-rays overlooked
by residents (Myles-Worsley, Johnston, & Simons, 1988), and
sommeliers can differentiate wines that simply taste “red” to the
uninitiated. If these experts had not learned to perceive the relevant
information in the stimulus array, their problem solving, judgment,
and abilities to learn from future experiences would be seriously
hampered.
We included the redrawing test on the prediction that T&P
would not help students notice the ratio structure in the clowns
worksheet, and therefore they would have no structures to remem-
ber when redrawing the worksheet. In addition, because they had
never learned to perceive ratio structures, they would not see them
embedded in the transfer tasks, and therefore they would fail to
transfer. Our prediction was based on the assumption that telling
first would lead T&P students to pay attention to what they had
just learned (the formulas) rather than search for the new infor-
mation in the worksheet (the physical ratio structure). In this
regard, T&P can sometimes exacerbate a more general phenome-
non, where prior knowledge filters out perceptual information, in
part, because people presume they have seen all that is necessary
to complete the task at hand (Neisser & Becklen, 1975; Nickerson
& Adams, 1979; Simons & Levin, 1998; Stevens & Hall, 1998).
With respect to the materials, the explicit use of contrasting
cases for learning found initial expression in J. J. Gibson and
Gibson’s (1955) demonstration that people learn to pick up infor-
mation in the environment by progressively noticing structure
across systematic variation. According to J. J. Gibson (1979), a
major task for psychology involves identifying the physical infor-
mation that makes perception possible. This differs from the cog-
nitive enterprise of identifying the mental structures that enable
people to enrich or go beyond the available information (see also
J. J. Gibson & Gibson, 1955). Identifying the information that
makes perception possible is highly relevant to the design of
guided discovery lessons, where the goal is to have students induce
the structure of phenomena. If the cases used for instruction do not
include sufficient information, students will not be able to perceive
 COMPARING USES OF CONTRASTING CASES
the underlying structure. Perceptual theory provides guidance for
deciding what information to include in the examples.
Perception and induction depend on finding structure within
variability (e.g., Paas & Van Merriënboer, 1994). For the topic of
physics learning, the theory of a perceptual gradient helps specify
what variability to include in the examples. According to J. J.
Gibson (1979), gradients make perception possible. A simple
example of a gradient is how objects in the distance appear smaller
than nearby objects. Paintings capitalize on the receding-into-the-
distance gradient so that people can see relative distances and
object size on flat canvas. The gradient makes it possible to see
depth. Painting a single object would not suffice. Creating instruc-
tional examples is similar to creating a painting. Examples need to
include sufficient information for students to perceive the desired
property. If we take the example of density, a single instance does
not provide sufficient information for students to perceive density;
there needs to be a gradient of densities.
The theory of gradients led to the specific contrasting cases used
in this study. For the crowded clowns worksheet (see Figure 1), we
created a density gradient across three companies. We chose three
companies, instead of two, to specify the linear aspect of the
gradient. We also used the same context (clowns within buses) so
that students could more readily discern the gradient than if we had
used three different contexts (e.g., an example of clowns in buses,
an example of atoms in a sphere). For the students in the current
studies, it was also necessary to include information specifying a
second gradient. Students of this age also needed to notice the
proportionality of mass-to-volume for a given density. Therefore,
within each company, we created a gradient by varying the number
of buses and clowns for its given density.
Our methodology for designing the contrasting cases involved
identifying and creating the relevant gradients. It is a method for
deciding what similarities and differences to include in the cases.
It differs from asking students to explore or freely experiment in
the hopes they will generate the relevant gradient information. It
also differs from picking instructional examples based on other
considerations, such as everyday familiarity.
Although the materials were created to include the sufficient and
relevant gradients, the inventing directive was included so that the
students would search for the property we intended them to per-
ceive. The T&P condition demonstrated that the presence of in-
formation does not guarantee its uptake. Critically, the inventing
task was also designed so that students would develop a rule that
accounts for the gradients. In paintings, observers can see depth
because artists have included the relevant gradient information.
The artists, however, differ from casual observers, because they
also have a set of rules that enable them to produce the relevant
gradients. They understand—rather than only experience—the re-
lations among objects that produce the perception of depth (e.g.,
how much to change relative sizes, shapes, and positions to indi-
cate the desired depth). With the inventing task, we simultaneously
wanted students to pick up the gradient of density and to develop
an understanding of the invariant that produces the gradient (ratios
of mass to volume).
It would take more refined research to parcel the relative value
of perceptual and cognitive theories for describing the learning in
the current studies. For example, it would be useful to determine
whether poor recall on the clown redrawing task is due to a failure
to perceive the ratios in the first place or due to a cognitive
15
problem such as missing the mental schema of ratio. It would also
be an interesting exercise to determine whether cognitive and
perceptual theories would make the same prescriptions for creating
the cases from which students learn. In the meantime, a major take
away from the current research is that it is not sufficient for
students only to hear and practice symbolic explanations, lest the
structure of the phenomenon gets lost in plain sight.
References
Atkinson, R. K., Derry, S. J., Renkl, A., & Wortham, D. (2000). Learning
from examples: Instructional principles from the worked examples re-
search. Review of Educational Research, 70, 181–214. doi:10.3102/
00346543070002181
Barron, B. J. S., Schwartz, D. L., Vye, N. J., Moore, A., Petrosino, A.,
Zech, L., . . . The Cognition and Technology Group at Vanderbilt.
(1998). Doing with understanding: Lessons from research on problem-
and project-based learning. Journal of the Learning Sciences, 7, 271–
311. doi:10.1207/s15327809jls0703&4_2
Barrows, H. S., & Tamblyn, R. M. (1976). An evaluation of problem-based
learning in small groups utilizing a simulated patient. Journal of Medical
Education, 51, 52–54.
Bassok, M., & Holyoak, K. J. (1989). Interdomain transfer between iso-
morphic topics in algebra and physics. Journal of Experimental Psy-
chology: Learning, Memory, and Cognition, 15, 153–166. doi:10.1037/
0278-7393.15.1.153
Biederman, I., & Shiffrar, M. M. (1987). Sexing day-old chicks: A case
study and expert systems analysis of a difficult perceptual-learning task.
Journal of Experimental Psychology: Learning, Memory, and Cogni-
tion, 13, 640 – 645. doi:10.1037/0278-7393.13.4.640
Bowers, J., Cobb, P., & McClain, K. (1999). The evolution of mathemat-
ical practices: A case study. Cognition and Instruction, 17, 25– 64.
doi:10.1207/s1532690xci1701_2
Bransford, J. D., Franks, J. J., Vye, N. J., & Sherwood, R. D. (1989). New
approaches to instruction: Because wisdom can’t be told. In S. Vosnia-
dou & A. Ortony (Eds.), Similarity and analogical reasoning (pp.
470 – 497). New York, NY: Cambridge University Press. doi:10.1017/
CBO9780511529863.022
Bransford, J. D., & Schwartz, D. L. (1999). Rethinking transfer: A simple
proposal with multiple implications. In A. Iran-Nejad & P. D. Pearson
(Eds.), Review of research in education (Vol. 24, pp. 61–101). Wash-
ington, DC: American Educational Research Association.
Brown, A. L., & Kane, M. J. (1988). Preschool children can learn to
transfer: Learning to learn and learning from example. Cognitive Psy-
chology, 20, 493–523. doi:10.1016/0010-0285(88)90014-X
Catrambone, R. (1998). The subgoal learning model: Creating better ex-
amples so that students can solve novel problems. Journal of Experi-
mental Psychology: General, 127, 355–376. doi:10.1037/0096-
3445.127.4.355
Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive
Psychology, 4, 55– 81. doi:10.1016/0010-0285(73)90004-2
Edelson, D. C., Gordin, D. N., & Pea, R. D. (1999). Addressing the
challenges of inquiry-based learning through technology and curriculum
design. Journal of the Learning Sciences, 8, 391– 450. doi:10.1207/
s15327809jls0803&4_3
Eisner, E. W. (1972). Educating artistic vision. New York, NY: Mac-
millan.
Fensham, P. J., & Kass, H. (1988). Inconsistent or discrepant events in
science instruction. Studies in Science Education, 15, 1–16. doi:10.1080/
03057268808559946
Forbus, K. (1984). Qualitative process theory. Artificial Intelligence, 24,
85–168. doi:10.1016/0004-3702(84)90038-9
Gentner, D., Loewenstein, J., & Thompson, L. (2003). Learning and
16 SCHWARTZ, CHASE, OPPEZZO, AND CHIN
transfer: A general role for analogical encoding. Journal of Educational
Psychology, 95, 393– 408. doi:10.1037/0022-0663.95.2.393
Gentner, D., & Markman, A. B. (1997). Structure mapping in analogy and
similarity. American Psychologist, 52, 45–56. doi:10.1037/0003-
066X.52.1.45
Gibson, E. J. (1969). Principles of perceptual learning and development.
New York, NY: Meredith.
Gibson, J. J. (1979). The ecological approach to visual perception. Boston,
MA: Houghton Mifflin.
Gibson, J. J., & Gibson, E. J. (1955). Perceptual learning: Differentiation
or enrichment. Psychological Review, 62, 32– 41. doi:10.1037/h0048826
Gick, M. L., & Holyoak, K. J. (1983). Schema induction and analogical
transfer. Cognitive Psychology, 15, 1–38. doi:10.1016/0010-
0285(83)90002-6
Gick, M. L., & Paterson, K. (1992). Do contrasting examples facilitate
schema acquisition and analogical transfer? Canadian Journal of Psy-
chology, 46, 539 –550. doi:10.1037/h0084333
Goldstone, R. L., & Sakamoto, Y. (2003). The transfer of abstract princi-
ples governing complex adaptive systems. Cognitive Psychology, 46,
414 – 466. doi:10.1016/S0010-0285(02)00519-4
Goldstone, R. L., & Son, J. Y. (2005). The transfer of scientific principles
using concrete and idealized simulations. Journal of the Learning Sci-
ences, 14, 69 –110. doi:10.1207/s15327809jls1401_4
Greeno, J. G., Smith, D. R., & Moore, J. L. (1993). Transfer of situated
learning. In D. K. Detterman & R. J. Sternberg (Eds.), Transfer on trial:
Intelligence, cognition, and instruction (pp. 99 –167). Norwood, NJ:
Ablex.
Gresalfi, M. S., & Cobb, P. (2006). Cultivating students’ discipline-specific
dispositions as a critical goal for pedagogy and equity. Pedagogies, 1,
49 –57. doi:10.1207/s15544818ped0101_8
Harp, S. F., & Mayer, R. E. (1998). How seductive details do their damage:
A theory of cognitive interest in science learning. Journal of Educational
Psychology, 90, 414 – 434. doi:10.1037/0022-0663.90.3.414
Hiebert, J., & Stigler, J. W. (2004). A world of difference: Classrooms
abroad provide lessons in teaching math and science. Journal of Staff
Development, 25, 10 –15.
Hiebert, J., Thomas, C. P., Fennema, E., Fuson, K., Human, P., Murray, H.,
. . . Wearne, D. (1996). Problem solving as a basis for reform in curric-
ulum and instruction: The case of mathematics. Educational Researcher,
25, 12–21. doi:10.3102/0013189X025004012
Hiebert, J., & Wearne, D. (1996). Instruction, understanding, and skill in
multidigit addition and subtraction. Cognition and Instruction, 14, 251–
283. doi:10.1207/s1532690xci1403_1
Hofstadter, D. H. (1995). Fluid concepts and creative analogies: Computer
models of the fundamental mechanisms of thought. New York, NY:
Basic Books.
Jee, B. D., Gentner, D., Forbus, K., Sageman, B., & Uttal, D. H. (2009).
Drawing on experience: Use of sketching to evaluate knowledge of
spatial scientific concepts. In N. A. Taatgen & H. van Rijn (Eds.),
Proceedings of the 31st Annual Conference of the Cognitive Science
Society (pp. 2499 –2504). Austin, TX: Cognitive Science Society.
Kaminski, J. A., Sloutsky, V. M., & Heckler, A. F. (2008). The advantage
of abstract examples in learning math. Science, 320, 454 – 455. doi:
10.1126/science.1154659
Kapur, M. (2008). Productive failure. Cognition and Instruction, 26, 379 –
424. doi:10.1080/07370000802212669
Kapur, M. (2010). Productive failure in mathematical problem solving.
Instructional Science, 38, 523–550. doi:10.1007/s11251-009-9093-x
Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why minimal
guidance during instruction does not work: An analysis of the failure
of constructivist, discovery, problem-based, experiential, and
inquiry-based teaching. Educational Psychologist, 41, 75– 86. doi:
10.1207/s15326985ep4102_1
Klahr, D., & Nigam, M. (2004). The equivalence of learning paths in early
science instruction: Effects of direct instruction and discovery learning.
Psychological Science, 15, 661– 667. doi:10.1111/j.0956-7976.2004
.00737.x
Lehrer, R., & Schauble, L. (2004). Modeling natural variation through
distribution. American Educational Research Journal, 41, 635– 679.
doi:10.3102/00028312041003635
Lesh, R. A., & Doerr, H. M. (Eds.). (2003). Beyond constructivism: Models
and modeling perspectives on mathematics problem solving, learning,
and teaching. Mahwah, NJ: Erlbaum.
Limón, M. (2001). On the cognitive conflict as an instructional strategy for
conceptual change: A critical appraisal. Learning and Instruction, 11,
357–380. doi:10.1016/S0959-4752(00)00037-2
Mayer, R. E. (2009). Constructivism as a theory of learning versus con-
structivism as a prescription for instruction. In S. Tobias & T. M. Duffy
(Eds.), Constructivist instruction: Success or failure (pp. 184 –200).
New York, NY: Routledge.
McNeil, N. M., Uttal, D. H., Jarvin, L., & Sternberg, R. J. (2009). Should
you show me the money? Concrete objects both hurt and help perfor-
mance on mathematics problems. Learning and Instruction, 19, 171–
184. doi:10.1016/j.learninstruc.2008.03.005
Michael, A. L., Klee, T., Bransford, J. D., & Warren, S. (1993).
The transition from theory to therapy: Test of two instructional
methods. Applied Cognitive Psychology, 7, 139 –153. doi:10.1002/
acp.2350070206
Myles-Worsley, M., Johnston, W. A., & Simons, M. A. (1988). The
influence of expertise on X-ray image processing. Journal of Experi-
mental Psychology: Learning, Memory, and Cognition, 14, 553–557.
doi:10.1037/0278-7393.14.3.553
Needham, D. R., & Begg, I. M. (1991). Problem-oriented training pro-
motes spontaneous analogical transfer: Memory-oriented training pro-
motes memory for training. Memory & Cognition, 19, 543–557. doi:
10.3758/BF03197150
Neisser, U., & Becklen, R. (1975). Selective looking: Attending to visually
specified events. Cognitive Psychology, 7, 480 – 494. doi:10.1016/0010-
0285(75)90019-5
Nickerson, R. S., & Adams, M. J. (1979). Long-term memory for a
common object. Cognitive Psychology, 11, 287–307. doi:10.1016/0010-
0285(79)90013-6
Paas, F. G. W. C., & Van Merriënboer, J. J. G. (1994). Variability of
worked examples and transfer of geometrical problem-solving skills: A
cognitive-load approach. Journal of Educational Psychology, 86, 122–
133. doi:10.1037/0022-0663.86.1.122
Rittle-Johnson, B., & Star, J. R. (2007). Does comparing solution methods
facilitate conceptual and procedural knowledge: An experimental study
on learning to solve equations. Journal of Educational Psychology, 99,
561–574. doi:10.1037/0022-0663.99.3.561
Rittle-Johnson, B., & Star, J. R. (2009). Compared with what? The effects
of different comparisons on conceptual knowledge and procedural flex-
ibility for equation solving. Journal of Educational Psychology, 101,
529 –544. doi:10.1037/a0014224
Roll, I. (2009). Structured invention activities to prepare students for future
learning: Means, mechanisms, and cognitive processes (Doctoral dis-
sertation). Carnegie Mellon University, Pittsburgh, PA.
Rothkopf, E. Z., & Billington, M. J. (1979). Goal guided learning for text:
Inferring a descriptive processing model from inspection times and eye
movements. Journal of Educational Psychology, 71, 310 –327. doi:
10.1037/0022-0663.71.3.310
Schmidt, H. G., De Volder, M. L., De Grave, W. S., Moust, J. H. C., &
Patel, V. L. (1989). Explanatory models in the processing of science
text: The role of prior knowledge activation through small-group dis-
cussion. Journal of Educational Psychology, 81, 610 – 619. doi:10.1037/
0022-0663.81.4.610
Schwartz, D. L., & Bransford, J. D. (1998). A time for telling. Cognition
and Instruction, 16, 475–522. doi:10.1207/s1532690xci1604_4
 COMPARING USES OF CONTRASTING CASES
Schwartz, D. L., Chase, C. C., & Bransford, J. D. (in press). Adaptive
transfer: The tension of routines and novelty. Educational Psychologist.
Schwartz, D. L., Lindgren, R., & Lewis, S. (2009). Constructivism in an
age of non-constructivist assessments. In S. Tobias & T. M. Duffy
(Eds.), Constructivist instruction: Success or failure (pp. 34 – 61). New
York, NY: Routledge.
Schwartz, D. L., & Martin, T. (2004). Inventing to prepare for future
learning: The hidden efficiency of encouraging original student produc-
tion in statistics instruction. Cognition and Instruction, 22, 129 –184.
doi:10.1207/s1532690xci2202_1
Schwartz, D. L., Martin, T., & Pfaffman, J. (2005). How mathematics
propels the development of physical knowledge. Journal of Cognition
and Development, 6, 65– 88. doi:10.1207/s15327647jcd0601_5
Siegler, R. S. (1981). Developmental sequences within and between con-
cepts. Monographs of the Society for Research in Child Development,
46(1, Serial No. 189).
Simons, D. J., & Levin, D. T. (1998). Failure to detect changes to people
in a real-world interaction. Psychonomic Bulletin & Review, 5, 644 –
649. doi:10.3758/BF03208840
Stevens, R., & Hall, R. (1998). Disciplined perception: Learning to see in
17
technoscience. In M. Lampert & M. L. Blunk (Eds.), Talking mathe-
matics in school: Studies of teaching and learning (pp. 107–148).
Cambridge, England: Cambridge University Press.
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning.
Cognitive Science, 12, 257–285. doi:10.1207/s15516709cog1202_4
Taylor, J. L., Smith, K. M., van Stolk, A. P., & Spiegelman, G. B. (2010).
Using invention to change how students tackle problems. CBE—Life
Sciences Education, 9, 504 –512. doi:10.1187/cbe.10-02-0012
Tobias, S., & Duffy, T. M. (Eds.). (2009). Constructivist instruction:
Success or failure. New York, NY: Routledge.
Tuovinen, J. E., & Sweller, J. (1999). A comparison of cognitive load
associated with discovery learning and worked examples. Journal of
Educational Psychology, 91, 334 –341. doi:10.1037/0022-0663.91.2.334
Williams, J. J., & Lombrozo, T. (2010). The role of explanation in
discovery and generalization: Evidence from category learning. Cogni-
tive Science, 34, 776 – 806. doi:10.1111/j.1551-6709.2010.01113.x
Received January 15, 2010
Revision received July 20, 2011
Accepted July 20, 2011 䡲